An automated valuation model (AVM) is software that estimates a property’s market value in seconds by running public records, recent sales and local market data through statistical and machine learning models. So if you are wondering what is AVM in real estate, the short answer is that it is the engine behind the instant home values you see on Zillow, Redfin and your lender’s portal, and more and more often it sits behind real lending decisions in the US too.<\/p>\n
That second part is why AVMs deserve more attention than they usually get. Since October 1, 2025, mortgage originators and secondary market issuers that use AVMs in covered credit decisions have had to follow federal quality control standards. Lenders now approve some home equity lines and refinances with no appraiser on site at all, and a growing number of proptech and fintech startups are building valuation features straight into their products.<\/p>\n
This guide walks through what an automated valuation model is, what goes into one, how AI has changed the way property gets priced, how accurate the numbers really are, and how AVMs compare with RVMs, CMAs and full appraisals. If you are a founder or technical lead thinking about building valuation into your own platform, the later sections cover the build steps and the options for getting it done.<\/p>\n
AVM stands for automated valuation model. It is a computer model that produces an estimated market value for a specific property as of a specific date, without anyone visiting the home. The estimate usually arrives with a value range and a confidence score so the person using it knows how much to trust the number.<\/p>\n
Most AVMs in the US are built on two classic techniques, often blended together.<\/p>\n
A third family, comparables based AVMs, picks a set of similar nearby sales for each property and adjusts them for differences, much like an appraiser would. Because the comps are chosen per property, the result can be traced back and explained, which lenders appreciate.<\/p>\n
Commercial grade AVMs rarely rely on a single method. They run several models, weight each result by how confident it is, and report one blended estimate. Freddie Mac’s Home Value Explorer, for example, has long combined a repeat sales model with a hedonic model to arrive at its final value.<\/p>\n
A typical AVM report includes an estimated value, a high and low range, a confidence score, the effective date and the comparable sales the model leaned on. The inputs behind that report usually come from five places.<\/p>\n
What an AVM does not include matters just as much. It has no idea that the kitchen was redone last year, that the roof leaks or that the house backs onto a busy road unless that information shows up somewhere in the data. That gap is the main reason AVM estimates can drift away from what a buyer would actually pay.<\/p>\n
Ten years ago, the honest answer to what is AVM in real estate would have been a regression formula sitting on top of county records. That is no longer the full picture. The best AVMs on the market now use machine learning at almost every step, and the gains come mostly from reading data that older models simply ignored.<\/p>\n
AI does not remove every weakness, though. A home that has not been listed in years has no fresh photos or description to read, so off market estimates still lean on older data. Machine learning can also absorb historical bias from the sales it learns from, which is part of why regulators now expect lenders to test AVMs for fair lending problems. Teams adding valuation to a product should plan for explainability from day one, and the same thinking applies if you are wiring AI agents into your workflows<\/a> around those valuations.<\/p>\n Automated valuation models grew out of academic pricing research in the 1970s and 1980s and moved into US mortgage lending through the 1990s. The table below traces the key moments, from the theory behind them to the federal rule that now governs their use.<\/p>\nWhen Did Automated Valuation Models Emerge?<\/h2>\n